Feature Overview

HyperShift brings about a new architectural pattern by decoupling the control-plane and data-planes of a cluster. This decoupling inherently adds a very important characteristic, that is, centralized management of control-planes. While this has numerous benefits as stated in the market problem [1], it also comes with a set of challenges, especially when it comes to tracking cost-per-tenant, and understanding the resource-consumption model per tenant.

With the current standalone model of HyperShift, a cluster resource-consumption model is easy to account for, since the CP and DP are in close-proximity it is more straightforward to estimate infrastructure costs, for the resources required by the Data-plane and the control-plane.

That said, it becomes important to bake-in the necessary hooks make metering and cost estimation a design principle rather than an afterthought.

Background, and strategic fit

Understanding the resource-consumption model and how it reflects back on costs is very important at scale and will likely be a demand for our larger customers.

Goal(s)

The goal of this feature is to provide means to:

• Track resource consumption of control-planes as a single entity in the management cluster namespace.
• Report costs per control-plane per cluster.

There are existing tools that can be utilized to achieve the desired outcome, for example, for tracking costs, there is KubeCost [3]. For metering, there is the OpenShift metering operator [4] (which is no longer maintained). 

References

[1] https://issues.redhat.com/browse/OCPPLAN-3783 

[2] https://github.com/kube-reporting/metering-operator

[3] https://kubecost.com/

[4] https://github.com/kube-reporting/metering-operator